ARTICLE

https://doi.org/10.1038/s41467-021-23595-x OPEN superfamily member 1 undergoes stimulus-coupled conformational reorganization to regulate metabolism in mice

Yue Li 1,2, Norihiro Imai3, Hayley T. Nicholls 3, Blaine R. Roberts 4, Samaksh Goyal 1,2, Tibor I. Krisko 3, Lay-Hong Ang 1,2, Matthew C. Tillman4, Anne M. Roberts4, Mahnoor Baqai 1,2, Eric A. Ortlund 4, ✉ ✉ David E. Cohen 3 & Susan J. Hagen 1,2 1234567890():,; In brown , thermogenesis is suppressed by thioesterase superfamily member 1 (Them1), a long chain fatty acyl-CoA thioesterase. Them1 is highly upregulated by cold ambient temperature, where it reduces availability and limits thermogenesis. Here, we show that Them1 regulates metabolism by undergoing conformational changes in response to β-adrenergic stimulation that alter Them1 intracellular distribution. Them1 forms metabolically active puncta near droplets and mitochondria. Upon stimulation, Them1 is phosphorylated at the N-terminus, inhibiting puncta formation and activity and resulting in a diffuse intracellular localization. We show by correlative light and electron microscopy that Them1 puncta are biomolecular condensates that are inhibited by phosphorylation. Thus, Them1 forms intracellular biomolecular condensates that limit fatty acid oxidation and sup- press thermogenesis. During a period of energy demand, the condensates are disrupted by phosphorylation to allow for maximal thermogenesis. The stimulus-coupled reorganization of Them1 provides fine-tuning of thermogenesis and energy expenditure.

1 Division of General Surgery, Department of Surgery, Beth Israel Deaconess Medical Center, Boston, MA, USA. 2 Department of Surgery, Harvard Medical School, Boston, MA, USA. 3 Division of Gastroenterology and Hepatology, Department of Medicine, Weill Cornell Medical College, New York, NY, USA. ✉ 4 Department of Biochemistry, Emory University, Atlanta, GA, USA. email: [email protected]; [email protected]

NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x

ntracellular triglycerides are stored within lipid droplets (LD) Them1 is comprised of tandem N-terminal thioesterase Ithat are juxtaposed to mitochondria in the cytoplasm1. This domains and a C-terminal lipid-binding steroidogenic acute close relationship facilitates the rapid transfer of fatty acids to regulatory-related lipid transfer (START) domain. Proteomics mitochondria, which are generated by lipolysis in response to has revealed that there are a number of BAT-selective phos- cellular energy demands2. One such demand is non-shivering phorylation sites near the N-terminus of Them1 in vivo, most thermogenesis, which is mediated by brown adipose tissue (BAT) notable are serine residues at (S) 15, 18, and 2512. Here, we show in response to cold exposure to generate heat3. This occurs, at experimentally that Them1 is phosphorylated at the N-terminus least in part, when mitochondrial β-oxidation is uncoupled from after stimulation in vitro and exists in two different oxidative phosphorylation via uncoupling protein 1. The hydro- phosphorylation-dependent conformational states: punctate and lysis of LD-derived triglycerides is initiated following the release diffuse. Punctate refers to small, highly concentrated aggregations of norepinephrine from neurons, which binds to β3 adrenergic of Them1 that localize to regions of the cytoplasm that interface receptors on brown adipocytes to stimulate a cell signaling cas- with LD and mitochondria, whereas diffuse refers to homo- cade that activates adenylyl cyclase to generate cAMP and activate geneously distributed Them1 across the cell cytoplasm and protein kinase A (PKA)4. In brown adipocytes, the generation of nucleus. We demonstrate that β-adrenergic stimulation with NE, fatty acids for mitochondrial oxidation requires PKA to phos- which is used to mimic cold exposure, activates a signaling cas- phorylate/activate perilipin, a regulatory membrane-bound pro- cade that results in Them1 phosphorylation and a diffuse loca- tein that encircles LD. The function of perilipin is to protect lization. A functional analysis revealed that Them1 in punctate triglycerides within the LD from lipolysis, and its phosphoryla- form suppresses fatty acid oxidation, whereas this suppression is tion by PKA allows access of cytoplasmic adipose triglyceride abrogated when it is diffusely distributed in the cell cytoplasm. (ATGL), hormone-sensitive lipase (HSL), and mono- Overall, these findings highlight the importance of Them1 acylglycerol lipase (MAGL), to generate free fatty acids through phosphorylation in the regulation of thermogenesis in BAT and sequential lipolytic steps5. Once fatty acids are free in the cyto- lend support for targeting Them1 for the management of - plasm, they are esterified with (CoA) by long-chain related disorders. acyl-CoA synthetase 1 (ACSL1) to form fatty acyl-CoAs, which are transported into mitochondria via carnitine palmitoyl trans- Results ferase 1 and metabolized to produce heat6. Fatty acyl-CoA N-terminal serine phosphorylation of Them1. To assess the molecules can also be hydrolyzed to fatty acids in the cytoplasm BAT-selective phosphorylation of Them1, we consulted the by acyl-CoA thioesterase (Acot) isoforms7,8. The fatty acids phosphomouse database12, which revealed phosphopeptides generated can be utilized to make additional fatty acyl-CoAs that containing phosphorylation events at serine (S) 15, 18, and 25 in are transported into mitochondria or can be returned to the LD the N-terminus of Them1 in vivo within BAT, which are highly for storage. conserved (Supplementary Fig. 1). Thioesterase superfamily member 1 (Them1), which is also To explore the role of Them1 S-phosphorylation in regulating known as brown fat inducible thioesterase or steroidogenic acute metabolism, cultured immortalized brown adipose cells (iBAs) regulatory protein-related lipid transfer (START) domain 14 from mouse BAT were used as an in vitro model. Because iBAs (StarD14)/Acot11, plays an important role in energy did not express Them1 mRNA or protein, as was the case for homeostasis9,10. Mice with the genetic deletion of Them1 exhibit primary cultured brown adipocytes11, Them1 expression required increased energy expenditure, which results in reduced weight plasmid or adenoviral transfection (Supplementary Fig. 2a–c) to gain when challenged with a high fat diet despite high food study its biochemical and physiological characteristics in vitro. To consumption10. This is attributable to increased mitochondrial examine whether Them1 is S-phosphorylated after stimulation, fatty acid oxidation, and mechanistic studies have demonstrated we determined the aggregate abundance of phosphopeptides that Them1 functions to suppress energy expenditure by limiting in the N-terminus of Them1 by mass spectrometry after stimulation triglyceride hydrolysis in BAT, thus inhibiting the mitochondrial with phorbol 12-myristate 13-acetate (PMA; Fig. 1a, b) normalized oxidation of LD-derived fatty acids10,11. to hormone-sensitive lipase as a housekeeping phosphopeptide

Fig. 1 Regulation of Them1 phosphorylation and its subcellular localization in iBAs. a LC-MS/MS data for iBAs expressing Ad-Them1-EGFP stimulated with PMA for 0–4 h. The data are presented as normalized aggregate abundance of N-terminal phosphopeptides using hormore sensitive lipase as a reference for normalization, which does not change after PMA stimulation (see Supplementary Fig. 3). Regression line indicates a positive and significant correlation between phosphorylation events at the N-terminus and time after stimulation. Data are means ± SE for n = 3 different experiments/timepoint. Statistical significance was determined by ANOVA on the regression line, where P < 0.001. Normality test passed (P = 0.759) and constant variance test, via Spearman Rank Correlation, passed (P = 0.460). b Schematic diagram of the experimental design in a. c N-terminal sequence of Them1 showing specific phosphorylation events at S25, S27, and S31 in the Them1 sequence (blue asterisks) as determined by LC-MS/MS. Phosphorylation events in vivo at S15, S18, and S25 are represented by red font.

2 NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x ARTICLE that does not change with PMA stimulation (Supplementary not attributable to the synthesis of new Them1-EGFP protein, Fig. 3). This resulted in a linear increase in phosphorylation because inhibition of protein synthesis using cycloheximide events over time after stimulation (Fig. 1a), with serine during the 4 h period of stimulation had no effect on Them1 phosphorylation events that could be measured at S25, S27, localization or protein concentration (Supplementary Fig. 7). and S31 (Fig. 1c). Taken together, these in vivo and in vitro The canonical PKC binding sequence, R/K-X-S-X-R/K where results suggested that stimulus-mediated S-phosphorylation Xat+1 is a hydrophobic residue, is not present within the N- within the N-terminus could control Them1 metabolic activity terminal Them1 sequence (Fig. 1c). However, our pathway in BAT. suggested that PKC was involved in the dissolution of puncta. To examine this experimentally, we used an antibody specific for the canonical PKC binding sequence, which showed that no antibody Phosphorylation-dependent changes in Them1 localization recognition of PKC-mediated S-phosphorylated Them1 occurred after stimulation. In transfected cells, Them1 exhibited a punc- (Supplementary Fig. 8). Instead, at least six proteins in iBAs were tate intracellular localization (Fig. 2a). When cells co-stained for time-dependently phosphorylated by PKC after PMA stimulation LD and mitochondria were reconstructed from 3-D stacks, (Supplementary Fig. 8). These results support that Them1 per se Them1 was closely associated, but did not co-localize, with either is not a PKC substrate but that PKC may be indirectly involved in of these organelles (Fig. 2b). The punctate localization of Them1 the dissolution of puncta after PMA stimulation (Fig. 2c). was not due to EGFP aggregation, per se, as evidenced by the We next sought evidence that the stimulus-coupled changes in diffuse intracellular localization of EGFP in cells transfected with Them1 localization also occurred in brown adipose cells in vivo EGFP alone (Supplementary Fig. 4a). Furthermore, Them1 (Fig. 3). For this, mice were injected with saline (control) or with formed puncta in transfected cells that co-localized with Them1- the β3 adrenergic receptor selective agonist CL316,243. In saline- EGFP (Supplementary Fig. 4b), and puncta were also present injected mice, Them1 localized to distinct puncta near LD and when Them1 was transfected without the EFGP tag (Supple- were in the cytoplasm (Fig. 3a). The strong fluorescence signal for mentary Fig. 4c). The antibody used to detect Them1 was made Them1 at baseline was due to mild cold exposure by housing and characterized by us11,13 and the newly prepared antibody, mice at 22 °C, which is below their thermoneutral zone. Tissues at which was also affinity purified using a Them1 peptide fragment, 4 h after saline injection showed the same localization of Them1 was specific for Them1 in BAT and iBAs in immunoblots (Sup- in puncta near lipid droplets and within the cytoplasm; Them1- plementary Fig. 5a) and specifically labeled structures in BAT containing puncta at 4 h after saline injection occupied a small from wild-type but not in Them1-deficient mice (Supplementary intracellular volume (Fig. 3b, c). Fig. 5b–d). By 1 h after CL316,243 administration, LD were reduced in size We next explored the hypothesis that the stimulus-mediated and number due to lipid hydrolysis and Them1 showed a larger phosphorylation of Them1 facilitates changes in its localization. cytoplasmic volume (Fig. 3c, d). By 2 h after CL administration, When differentiated iBAs were incubated with PMA, which LD were less abundant and Them1 occupied a larger cytoplasmic activates protein kinase C (PKC), and the localization of Them1 volume compared to 1 h (Fig. 3c, e). By 4 h after CL316,243 was examined in fixed or living cells, PMA initiated the administration, there were few intracellular LD, as evidenced by dissolution of Them1 from puncta (Fig. 2c, d). The dissolution the loss and disruption of perilipin 1 expression, and Them1 of puncta resulted in Them1 uniformly distributed within the cell occupied a larger cytoplasmic volume yet (Fig. 3c, f). To cytoplasm (Fig. 2d), which occurred over a 4-h period after PMA accompany the diffuse localization of Them1 from 1–4 h after stimulation (Fig. 2e) but not with vehicle alone (Supplementary CL administration was an increase in the nuclear localization of Fig. 6), as determined by using time-lapse live-cell microscopy. Them1 (Fig. 3d–f). When taken together with observations We further observed that the isoform-specific PKCβ inhibitor in vitro using iBAs, these in vivo studies support the notion that ruboxistaurin (LY333531 or LY) completely blocked the dissolu- reorganization of Them1 puncta occurs after stimulation that tion of puncta in PMA-treated cells (Fig. 2d). activates the β3 adrenergic receptor (Fig. 2c). We next sought to identify upstream signaling events and define the pathway leading to Them1 phosphorylation (Fig. 2c). iBAs incubated with the PKA inhibitor PKI or the ATGL S-phosphorylation of the N-terminus of Them1 drives locali- inhibitor atglistatin prior to stimulation with PMA had a diffuse zation after stimulation. To develop direct evidence that phos- Them1 distribution in the cytoplasm (Fig. 2f). Similarly, iBAs phorylation of N-terminal serine residues of Them1 are involved stimulated with forskolin (Fig. 2g), which activates adenyl cyclase in puncta formation and dissolution, we deleted the first 36 and protein kinase A (PKA) upstream of PKC (Fig. 2c), or with amino acids of Them1 to generate a construct (Δ1-36-EGFP) that norepinephrine, which activates both PKA and PKC (Fig. 2c) did not contain the phosphorylation sites that were detected resulted in diffusely localized Them1, which was blocked by in vivo or following transfection and stimulation of iBAs (Fig. 4a, inhibiting downstream effectors with PKI, Atglistatin, or LY, b). The Δ1-36 mutant of Them1 exhibited a diffuse localization respectively (Fig. 2g, h). DAG released by ATGL is in the sn-1,3 (Fig. 4c). We then linked a peptide containing the first 36 amino or sn-2,3 stereo isoform, so that triacylglycerol-derived DAG acids of Them1 directly to EGFP (Fig. 4b), which exhibited the cannot activate PKC without isomerization to sn-1,2 DAG14. This same punctate localization as full-length Them1 (Fig. 4c). These event was not investigated, but would be required to confirm the results strongly support that amino acids 1–36 direct Them1 pathway details (Fig. 2c). Overall, these results suggest that localization to puncta and suggest that phosphorylation of Them1 norepinephrine stimulation activates PKA, and the activation of at one or more serine residues in the N-terminal region drives the PKC occurs downstream of PKA activation (Fig. 2c). Measuring conformational change in Them1 after stimulation. the volume of intracellular Them1 in iBAs transfected with To further explore this idea, we mutated each putative Them1-EGFP with or without inhibitors (Fig. 2i, schematic; green phosphorylation site by exchanging S for aspartic acid (S15-18- EFGP fluorescence signal) further demonstrated that Them1 in 25D, or DDD), which mimicked the phosphorylated state of the S puncta is concentrated and occupies significantly less volume amino acid residues that are phosphorylated in vivo (Fig. 4b). than occurs when Them1-EGFP is diffusely distributed after This construct resulted in diffuse Them1 localization (Fig. 4d). stimulation with PMA or forskolin (Fig. 2i). The dissolution of When the S sites were mutated in pairs (Fig. 4e), or individually Them1 from puncta after stimulation with PMA or forskolin was (Fig. 4f), S15D and S25D led to diffuse Them1 localization.

NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x

Fig. 2 Regulation of Them1 subcellular localization in iBAs. a Confocal microscopy of an iBAs expressing Them1-EGFP (green). LD lipid droplets (magenta), N nucleus (blue), P puncta (green). b 3-D image of an iBAs expressing Them1-EGFP (green), perilipin 1 to identify lipid droplets (LD, magenta), and Tomm20 to identify mitochondria (M, red). This image is shown in surface projection mode. c Schematic illustration of the putative Them1 regulatory pathway required for the phosphorylation and 3-D organization of Them1. β3-AR beta-3 adrenergic receptor, DAG diacylglyerol, NE norepinephrine. DAG and PKC may have an indirect effect on puncta dissolution (dashed lines). d PMA-induced activation of iBAs expressing Them1-EGFP (green) resulted in the diffusion of Them1, which can be blocked by the PKCβ inhibitor LY333531 (LY). e Live-cell imaging was used to record changes in Them1 localization over time after PMA treatment. LD lipid droplets, P1–P3 individual puncta (green). f After PMA-induced PKC activation, neither PKI or atglistatin (Atglst) blocked the diffusion of Them1 from puncta. g, h Whereas the activation of PKA and PKC by forskolin (FSK) or norepinephrine (NE) resulted in the diffusion of Them1, PKI, Atglst, and LY blocked the diffusion of puncta (P, green). Scale bars in a–h,10μm. i Box and whisker plot with individual cell data (green squares). For each box, a solid line represents the median and a dashed line is the mean. Outliers are shown at the 5th/95th percentiles from 3 to 5 different experiments with the following n: Them1-EGFP, n = 16; PMA, n = 8; PMA + PKI, n = 16; PMA + Atglst, n = 7; PMA + LY333531, n = 12; FSK, n = 9; FSK + PKI, n = 12; FSK + Atglst, n = 7; FSK + LY333531, n = 10. For PMA, *P = 0.024 PMA + LY vs PMA alone; †P < 0.001 PMA + LY vs PMA + PKI; ‡P < 0.01 PMA + LY vs PMA + Atglst. For forskolin, *P < 0.001 forskolin + PKI vs forskolin alone; †P = 0.011 forskolin + Atglst vs forskolin alone; ‡P = 0.006 forskolin + LY vs forskolin alone. ND no difference. Statistical significance as determined by a multiple comparisons ad hoc test (Dunns Method) after Kruskal–Wallis one-way ANOVA on ranks.

4 NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x ARTICLE

This result was quantified as the volume of Them1 per total cell Puncta formed by Them1 are metabolically active structures. volume using the strategy in Fig. 2i (Fig. 4g). By contrast, Although plasmid-mediated transfection of Them1 constructs exchanging S for alanine at S15 or at S15, S18, and S25 (AAA, was sufficient for fluorescence-based imaging studies, the low Fig. 4b), a configuration that cannot be phosphorylated, resulted transfection efficiencies prevented quantitative assessments of the in the punctate localization of Them1 alone or after stimulation impact of puncta formation on cellular metabolism. To obtain with PMA, FSK, or norepinephrine (Fig. 4h). These results higher transfection efficiency, we constructed adenoviral vectors demonstrate that the phosphorylation of Them1 at S15 and/or (Ad). These included wild-type Them1 or Them1 with AAA or S25 results in the diffuse localization of Them1 in iBAs. DDD substitutions and EGFP fused at the carboxy terminus

NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x

Fig. 3 Them1-containing puncta in brown adipose tissue (BAT) diffuse after stimulation in vivo. a BAT was excised from mice immediately after the administration of saline. These tissues consisted of large lipid droplets (LD; magenta) surrounded by perilipin and numerous Them1-containing puncta (P; green) that were localized in the cytoplasm or in association with perilipin at the lateral edge of LDs. N, nuclei (blue) were stained with Hoechst 33342 (HOE). Scale bar, 10 μm. n = 5 images taken randomly from n = 2 mice with similar results. b BAT tissues from mice injected with saline and then tissues collected 4 h later. The localization of lipid droplets (LD, magenta), puncta (P, green), and nuclei (N, blue) were the same as in a. n = 5 images taken randomly from n = 2 mice with similar results. Scale bar, 10 μm. c Quantification of the 3-D volume of Them1 from mice injected with saline and collected 4 h later (contr; control) or CL316,243, a β3 adrenergic receptor agonist, at 1, 2, or 4 h after stimulation. Regression line indicates a positive and significant correlation between the volume of Them1 and time; the volume of Them1 is small when Them1 localizes to puncta in control cells and large as Them1 is diffuse after stimulation with CL316,243 (see Fig. 2i). Error bars represent mean ± SE from n = 5 images taken randomly from n = 2 mice/treatment and analyzed using Imaris software. Statistical significance was determined by ANOVA on the regression line, where P = 0.006. d BAT tissues from mice exposed to CL316,243 for 1 h. In these tissues, LD (magenta) were smaller than in saline-treated mice and Them1 (green) was more diffuse. N nuclei (blue). e By 2 h after CL316,243 administration, LD were less abundant and Them1 occupied a larger cytoplasmic volume. f BAT tissue from mice 4 h after the administration of CL316,243. LD (magenta) were not present, perilipin was cytoplasmic, and Them1 (green) was nearly completely diffuse in the cytoplasm. Over time (compare d–f) nuclear (N) Them1 staining (blue) increased in intensity. d–f n = 5 images taken randomly from n = 2 mice/treatment with similar results. Scale bars, 10 μm.

(Fig. 4a, b) with Ad-EGFP used as a control. Them1 expression at Supplementary Fig. 9c). These results argue against downstream an MOI of 1:40 was ~45%, which resulted in Them1 expression regulation of these protein constructs by PMA. similar to tissue expression in BAT of cold-exposed mice (Sup- These metabolic studies support the role of Them1 puncta in plementary Fig. 2b). Both wild type (not shown) and AAA regulating oxidative metabolism in BAT, which were further (Fig. 5a) substituted Ad-Them1-EGFP formed puncta in iBAs. supported by measuring fatty acid oxidation contained within Using near super-resolution imaging via Zeiss Airyscan, it was endogenous triglycerides in iBAs by pulse labeling with 3H-oleate clear from 3-D images that Them1 resides in discrete puncta that and then determining triglyceride utilization during a chase were closely associated with LD (Fig. 5a). In contrast, DDD period (Supplementary Fig. 9d). During the pulse period, equal substituted Ad-Them1-EGFP exhibited a diffuse localization fatty acid uptake was observed in iBAs transduced with Ad- (Fig. 5b). In the near super-resolution 3-D images, diffuse Them1 EGFP, Ad-AAA-EGFP, or Ad-DDD-EGFP, leading to equal filled the entire volume of cytoplasm not occupied by cellular cellular accumulation of triglycerides (Supplementary Fig. 9d). By organelles (Fig. 5b). These results overall were consistent with the contrast, during the chase period triglyceride accumulated in iBAs results from our plasmid constructs. transduced with Ad-AAA-EGFP, but not Ad-EGFP or Ad-DDD- We next examined the role of Them1 in regulating the oxygen EGFP (Supplementary Fig. 9d), which reflected reduced rates of consumption rate (OCR) in differentiated iBAs in response to fatty acid oxidation in the presence of Ad-AAA-EGFP (Supple- norepinephrine stimulation. OCR values are a surrogate measure mentary Fig. 9e). of oxidation rates of fatty acids generated by hydrolysis of LD- derived triglycerides11. After optimizing conditions of cell density Puncta are biomolecular condensates, or “membraneless by calculating the total number of EGFP-containing cells/total organelles”, by correlative light/electron microscopy.To cells per plate (Supplementary Fig. 9a), OCR values were examine the ultrastructural details of puncta in iBAs, we next measured at baseline and after norepinephrine stimulation. used Them1-EGFP vector constructs that included an ascorbate Baseline and norepinephrine-stimulated OCR values were similar peroxidase–derived APEX2 tag appended to the carboxy terminus for Ad-DDD-EGFP and the Ad-EGFP control, whereas to perform correlative light/electron microscopy (Fig. 6). The norepinephrine-stimulated values of OCR for Ad-Them1-EGFP APEX2 tag was developed with diaminobenzadine, resulting in and Ad-AAA-EGFP were reduced (Fig. 5c). These findings brown reaction product only in cells expressing Them1 (Fig. 6a). suggest that Them1 is active in suppressing LD-derived FA This procedure allowed us to clearly determine which cells were oxidation when in its punctate configuration. Because the transfected and expressing Them1. Them1 puncta were imaged stimulated redistribution of Them1-EGFP-APEX2 in puncta fully using the AAA vector construct (AAA-EGFP-APEX2; Fig. 6b, c), diffuses over a period of 4 h (Fig. 2e), increased OCR values and diffuse Them1 was evaluated using the DDD vector construct within the first 1 h after norepinephrine exposure (Fig. 5c) would (DDD-EGFP-APEX2; Fig. 6d). Cells transfected with EGFP- be expected to reflect only partial dissolution of Them1- APEX2 without Them1 showed pale staining uniformly containing puncta. By pre-incubating iBAs expressing Ad- throughout cells including the nucleus (not shown). Them1-EGFP with PMA prior to norepinephrine exposure, Consistent with fluorescence images of iBAs transduced with which would result in the further dissolution of puncta, we Them1-EGFP (Fig. 2a), puncta in electron micrographs of iBAs observed an increase in OCR values relative to no PMA (Fig. 5d). transduced with AAA-EGFP-APEX2 were found exclusively in The addition of PMA to iBAs led to a progressive increase in the cytoplasm in close proximity to both LD and mitochondria values of extracellular acidification rate (ECAR) until the time of (Fig. 6a–c). Although the puncta resembled mitochondria from NE exposure in the absence of differences in OCR (Supplemen- steroid-secreting cells15, there was no apparent membrane tary Fig. 9b). This result suggests that PMA stimulated anaerobic surrounding the structure. Instead, puncta were an aggregate of glycolysis, as opposed to fatty acid oxidation, which increases small round and elongate droplets with amorphous boundaries OCR following NE exposure. The PMA-induced increases in (Fig. 6b, c). Diffusely localized Them1 in iBAs transduced with anaerobic glycolysis led to observed reductions in fatty acid DDD-EGFP-APEX2 was not clearly associated with any cyto- oxidation, as reflected by OCR in response to norepinephrine plasmic structure (Fig. 6d). stimulation, expressed as a percentage of baseline (Fig. 5e). When Because puncta were membraneless structures with character- the effects of PMA treatment on OCR per se are taken into istics similar to that described for biomolecular condensates account, the NE-stimulated suppression of OCR on cells within the cell cytoplasm16, we evaluated the Them1 primary transduced with Ad-AAA was maintained compared to cells sequence for characteristics that would facilitate a phase transduced with Ad-EGFP or Ad-DDD, (Fig. 5f and separation, aggregation, and/or the formation of puncta (Fig. 7).

6 NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x ARTICLE

Fig. 4 Phosphorylation at S15 and S25 in the N-terminus regulates the diffusion of Them1 in iBAs. a, b Illustration of the Them1 structure and its mutant constructs tagged with EGFP used in the experiments. a Them1, highlighting the N-terminus S phosphorylation sites, linked to EGFP at the carboxy terminus. b Δ1-36, deletion of amino acids 1-36 in the Them1 sequence; 1-36, a construct containing only the first 36 amino acids in the Them1 sequence; AAA, serine residues at amino acids 15, 18, and 25 were mutated to alanine (A); DDD, serine residues at amino acids 15, 18, and 25 were mutated to aspartic acid (D). All mutant Them1 constructs were linked to EGFP at the carboxy terminus. c the Δ1-36 Them1 construct diffuses in the cytoplasm (green) whereas the 1-36 construct forms puncta (green) around lipid droplets (LD), which were stained with Oil Red O stain (magenta). d Mutation of S15/S18/S25 to aspartic acid (DDD) results in diffuse Them1 localization (green). e S15/S18 or S15/S25 mutant constructs result in diffuse Them1 localization (green) whereas the S18/S25 construct expressed some puncta and some diffuse Them1 (green). f The single mutation at S15 or S25, but not S18, to aspartic acid resulted in the diffusion of Them1 (green). g Box and whisker plot with individual cell data (green squares) to quantify volume measurements for mutation of S15/S18/S25 (DDD), or single mutations at S15D, S18D, or S25D. For each box, a solid line represents the median and a dashed line is the mean. Outliers are shown at the 5th/95th percentiles. n = 2–5 cells each from two different preparations. Data were evaluated using a one-way ANOVA with all pairwise multiple comparison using the Holm–Sidak method. *P = 0.05, S18D vs S15D; †P = 0.011, S25D vs S18D. h The serine to alanine mutation at S15 results in Them1-containing puncta, which are not dispersed after activation by PMA, FSK, or NE. Scale bar in c–h,10μm. For each construct evaluated in Fig. 4, the data are representative of n = 3 independent experiments with 3–5 individual cells photographed per treatment with similar results.

NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x

Bioinformatic analysis of the amino acid sequence for Them1 that the N-terminal 20 amino acids were a highly disordered (Fig. 7a and Supplementary Fig. 10a) showed a highly disordered binding region (Fig. 7e). Interestingly, two of the phosphorylation region at the N-terminus spanning residues 20-45 as predicted by sites, S15 and S18, lie in this region (Fig. 7a), suggesting that the IUPRED (prediction of intrinsic disorder) curves (Fig. 7b). There phosphorylation sites may regulate binding events. Because were no prion-like regions in the sequence as determined by PLD multivalent interactions between proteins can lead to a phase (Fig. 7c) or FOLD curves (Supplementary Fig. 10b). The N- separation16, we also performed a motif scan using the Motif Scan terminus contains a high proportion of charged residues with a server (https://myhits.isb-sib.ch/cgi-bin/motif_scan). This analy- patch of basic residues followed by a patch of acidic residues sis highlighted no strong association of the 1st 20 amino acids (Supplementary Fig. 10c, d), which could engage in non-covalent with other proteins. crosslinking with other proteins or itself to drive a phase transition17. However, the PScore, a pi-pi interaction predictor, Discussion did not reach significance (Fig. 7d), which suggests that the Them1 was first identified as a that is markedly upregulated sequence of Them1 does not have the propensity to phase in response to cold ambient temperatures and is suppressed by separate based on pi–pi interactions. ANCHOR2 analysis, which warm temperatures9. Whereas this initially suggested that Them1 predicts protein binding regions in disordered proteins, showed function was to promote thermogenesis9, homozygous disruption

8 NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x ARTICLE

Fig. 5 The punctate localization of Them1 is metabolically active. a Ad-Them1-EGFP with S15, S18, and S25 mutated to alanine (Ad-AAA-EGFP, green) formed puncta near lipid droplets (LD, magenta). This close relationship was highlighted by using near super-resolution Zeiss Airyscan imaging and reconstructing cells in 3-D. b Phosphomimetic mutations at S15, S18, and S25 (Ad-DDD-EGFP, green) caused diffusion of Them1. In 3-D, Them1 filled the thickness of the cytoplasm (double white arrow) and could be found above and below the plane of organelles including LD (magenta), which excluded Them1. N nucleus (blue). Scale bars, 10 μm (low magnification images-left and center); 2.5 μm (high magnification image-right). Data are representative of n = 3 independent experiments with similar results. c iBAs transduced with Ad-Them1-EGFP (blue), Ad-AAA-EGFP (red), Ad-DDD-EGFP (violet), or Ad- EGFP (green) were stimulated with norepinephrine (NE). The response of OCR values to stimulation was measured as % of baseline. Data (n = 66–72 wells from three independent experiments) were evaluated using mixed-effects analysis (group effect P = 0.001; interaction effect P < 0.001) followed by multiple comparison testing using the Tukey method (*P = 0.015, 0.034, 0.041, 0.039, 0.030, Ad-Them1-EGFP vs. Ad-EGFP; †P = 0.031, 0.040, 0.034, 0.018, 0.013, 0.009, 0.007, Ad-AAA-EGFP vs. Ad-DDD-EGFP). d OCR values for Ad-Them1 (blue) with or without PMA. Data were evaluated using mixed-effects analysis (group effect P < 0.001; interaction effect P < 0.001) followed by multiple comparison testing using the Sidak method (*P = 0.012, <0.001, <0.001, <0.001, 0 vs. 3 μM PMA). e OCR values for Ad-EGFP (green), Ad-AAA-EGFP (red), and Ad-DDD-EGFP (violet) with or without PMA. Data were evaluated using mixed-effects analysis (vehicle: group effect P = 0.050, interaction effect P < 0.001; PMA: group effect P = 0.060, interaction effect P = 0.015) with multiple comparison testing using the Holm-Sidak method (vehicle: #P = 0.005, 0.003, Ad-EGFP vs. Ad-AAA-EGFP, *P = 0.034, Ad-DDD-EGFP vs. Ad-AAA-EGFP; PMA: #P = 0.032, 0.024, 0.024, Ad-EGFP vs. Ad-AAA-EGFP). f Change in area under the curve (Δ AUC) for vehicle – PMA. Data (n = 33–36 wells from three independent experiments) were evaluated using a one-way ANOVA (P = 0.004) with multiple comparison testing using the Holm–Sidak method (*P = 0.046, Ad-DDD-EGFP vs Ad-AAA-EGFP, ##P = 0.002, Ad-EGFP vs Ad-AAA-EGFP. Data are mean ± SE.

Fig. 6 Correlative light and electron microscopy identifies puncta as biomolecular condensates. a Thick, 0.5 μm, plastic sections of iBAs in culture transfected with phosphorylation mutants AAA- or DDD-Them1-EGFP-APEX2. Cells stained brown (box) also have dark brown punctate structures in mutant AAA-Them1 expressing cells (arrows). For AAA-transfected cells, serial sections show two different levels of the same cell. No puncta are present in cells transfected with the DDD-Them1 mutant although Them1 is found in the nucleus (arrows). Scale bar, 25 μm. b In thin sections of the boxed cell in a, puncta 1 (P1) is localized near the nucleus (N) and puncta 2 (P2) is localized near a lipid droplet (LD). At high magnification, punctate structures, P1 and P2, have no surrounding membrane and consist of liquid droplets embedded in an amorphous matrix. Scale bars, 5 μm (low magnification; original magnification, ×4000) and 0.5 μm (high magnification; original magnification, ×60,000). c In thin sections of the boxed cell in a, this section shows smaller puncta in the cytoplasm (P3), puncta associated with lipid droplets (LD; P4), and puncta in close association with LD and mitochondria (M) in the cell cytoplasm (P5). Scale bars, 5 μm (low magnification; original magnification, ×4000) and 0.5 μm (high magnification; original magnification, ×60,000). d In thin sections from the boxed region in a, no puncta are found in cells transfected with the DDD-Them1 mutant. Them1 is distributed throughout the cytoplasm and in the nucleus (N; arrows). Scale bars, 10 μm (low magnification; original magnification ×5000) and 1 μm (higher magnification; original magnification, ×25,000). All images are representative of the results obtained from n = 3 independent experiments. of Them1 in mice led to increased energy expenditure10,11. This temperatures, which also upregulates thermogenic ; Them1 result indicated that Them1 functions, on balance, to suppress is phosphorylated and inactivated by the same adrenergic sti- energy expenditure. The current study helps to explain how a mulus that promotes thermogenesis. Additionally, our studies suppressor of energy expenditure can be induced by cold ambient have demonstrated phosphorylation-dependent intracellular

NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x

Fig. 7 Bioinformatic analysis of Them1 suggests that a disordered region at the N-terminus spanning residues 20–45 may be involved in puncta formation. a Schematic diagram of the Them1 amino acid sequence aligned with the amino acid number from the N- to C-terminus. Thio indicates the position of thioesterase domains 1 and 2 and START indicates the position of the START domain. b Results from the IUPRED server, which predicts regions of disorder. The dashed line signifies the threshold for significance. The data indicate a large region of disorder in the first 50 amino acids (highlighted in gold). c Results from the Prion-like Amino Acid Composition server. The dashed line at 0.5 is a threshold for significance. The graph, red line, demonstrates no prion-like regions in the Them1 amino acid sequence. d Results from the PScore, which is a predictor of the propensity to phase separate based on pi–pi interactions. The dashed line is a threshold set at 4 to determine significance; the amino acid sequence of Them1 did not reach significance. e Results from an ANCHOR2 analysis, which predicts protein binding regions in disordered proteins. This analysis demonstrates that the N-terminal 20 amino acids of Them1 were strongly predicted to be a disordered binding region (highlighted in red). Two of the phosphorylation sites (red font) lie in this region (expanded region, also refer to Fig. 4b), suggesting they could potentially regulate a binding event. redistribution of Them1 in response to thermogenic stimuli, mitochondria and suggests that phosphorylation within the N- suggesting a model for the regulation of energy expenditure terminus amino acids 1–36, likely S15 phosphorylation, and to a within BAT. For this, Them1 activity is transiently silenced lesser degree S25 phosphorylation, relocates and inactivates during peak thermogenesis by relocation away from its active site Them1 in vivo. Levels of S15 phosphorylation that were below in puncta, which are located in close proximity to LD and detectable levels in our study could be due to the nature of mitochondria. peptide cleavage sites for mass spec analysis, glycosylation of Our prior observations indicate that the activity of Them1 in nearby sites, or indicate that PMA-induced stimulation is not brown adipocytes is to suppress the mitochondrial ß-oxidation of relevant to S15 phosphorylation, which may instead reflect nor- LD-derived fatty acids11. In keeping with this possibility, we epinephrine stimulation. These interesting possibilities must be observed that Them1 localized to puncta in the absence of pursued in future studies. Phosphopeptide analysis identified adrenergic stimulation both in vitro and in vivo. These structures other potential sites, including S27 and S31, the roles of which in are juxtaposed with mitochondria and LD such that they could be Them1 localization remain unknown but also need to be studied expected to regulate fatty acid trafficking between these two experimentally. Our inhibitor data suggest that PKCβ plays some organelles. Moreover, experiments in iBAs transduced with role in regulating Them1 metabolic function by changing the recombinant adenovirus revealed that full-length Them1 in physical state of Them1. This may be by phosphorylating one or unstimulated cells and the AAA-Them1 mutant, which is unre- more Them-1-interacting proteins after stimulation or other yet sponsive to phosphorylation, were localized to puncta and were undefined mechanisms. Interestingly, PKCβ appears to be active in suppressing norepinephrine-stimulated oxygen con- involved in coordinating thermogenesis, as evidenced by sumption. These results are similar to our previous findings in increased rates of fatty acid oxidation in both BAT and white cultured primary brown adipocytes, where Them1 actively sup- adipose tissue in mice globally lacking this kinase18. pressed norepinephrine-stimulated oxygen consumption11, The important role played by Them1-containing puncta in although puncta formation and dissolution were not investigated inhibiting the mitochondrial oxidation of LD-derived fatty acids in that study. begs the question about the composition of puncta, including By contrast, our data demonstrated that Them1 in its diffuse how they are formed and why they disappear after phosphor- configuration was ineffective at suppressing oxygen consumption. ylation. Our data using correlative light and electron microscopy This result was consistent with its relocation away from LD and suggest that Them1 constitutively resides in biomolecular

10 NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x ARTICLE condensates, also known as membraneless organelles, much like Together, our results reveal an unexpected mechanism for the the nucleoleus, stress granules, and p-bodies19,20. Biomolecular regulation of Them1, a protein that reduces fatty acid availability condensates form by liquid-liquid phase separation or for β-oxidation in mitochondria and limits thermogenesis. condensation21, which favors aggregation. The resulting structure Manipulation of Them1 localization in BAT may provide a often resembles an aggregation of round liquid droplets16, which therapeutic target for the management of metabolic disorders is the result we found here for Them1 in puncta by electron including obesity and non-alcoholic fatty liver disease. microscopy. The driving force underlying the formation of bio- molecular condensates is the exchange of macromolecular/mac- Methods romolecular and water/water interactions under conditions that Culture of brown adipocytes. Immortalized brown preadipocyte cells (iBAs) from are energetically favorable19. What these conditions might be for mouse brown adipose tissue25 were obtained from Dr. Bruce Spiegelman. iBAs ’ fi ’ Them1 are unknown but may be related to temperature, localized were cultured in Dulbecco s Modi ed Eagle s medium (DMEM; Gibco, Grand Island, NY), 20% fetal bovine serum (FBS; Gibco), 1% penicillin/streptomycin pH changes, or co-solutes that are expressed. Typical phase dia- solution (P/S; Gibco), and 2 μM HEPES (Gibco). Confluent iBAs were induced to grams can define the conditions needed for condensation, which differentiate from preadipocytes by adding induction cocktail containing 5 μM would be required to fully understand the localization of Them1 dexamethasone (Sigma-Aldrich, St. Louis, MO), 125 μM indomethacin (Sigma- μ μ to puncta in vitro and in vivo. Aldrich), 0.02 M (Sigma-Aldrich), 500 M isobutylmethylxanthine (Life Technologies, Carlsbad, CA), 1 nM triiodothyronine (Sigma-Aldrich), and 1 μM One consistent signature of biomolecular condensates is the rosiglitazone (Sigma-Aldrich) in medium consisting of DMEM, 10% FBS, and 1% concept that their formation includes scaffold and client P/S. After 48 h of induction, the medium was replaced by maintenance medium proteins19. Scaffold molecules are drivers of the condensation consisting of DMEM, 10% FBS, 1% P/S with 0.02 μM insulin, 1 nM triiodothyr- process and clients are molecules that partition into the con- onine, and 1 μM rosiglitazone. The maintenance medium was replaced every fi 2 days. For immunostaining and confocal applications, iBAs were plated on glass densate. It is not clear into which category Them1 ts, although coverslips, which were pre-treated with 0.01% collagen type I (Sigma-Aldrich) in the characteristics of its molecular structure suggest that it could 0.01 M acetic acid. be the primary scaffold protein. Phase separation of scaffold proteins and the subsequent partitioning of their clients are dri- Production of DNA constructs. The full-length cDNA for Them1 (from Mus ven by network interactions for which two distinct classes of musculus: Q8VHQ9) was cloned and linked to EGFP plus an engineered ascorbate protein architecture are thought to be involved19. The first is peroxidase (APEX2) fusion protein the C-terminus. Codons for S15, S18, and S25 were mutated to alanine (A) or aspartic acid (D) using the QuikChange muta- multiple folded (SH3) domains that interact with short linear Δ 19 genesis kit (Agilent, Santa Clara, CA). The 1-36 mutation includes deletion of the motifs . Since Them1 does not have an SH3 domain, as deter- first 36 amino acids at the N-terminus of Them1, whereby Them 1 begins at amino mined by using the Simple Modular Architecture Research Tool acid 37 (M; AUG). The full-length cDNA for Them1 was cloned into a pVQAd web server22, this could not be the sequence element driving CMV K-NpA vector (ViraQuest, North Liberty, IA) along with C-terminal EGFP puncta formation. Second is intrinsically disordered regions19, for downstream of the CMV promoter. The Them1 adenovirus was created by ViraQuest. which Them1 has a strong candidate region near the amino terminus; amino acids 20–45 represent a highly disordered area Heterologous Them1 expression in cultured iBAs. Plasmid transfection with that includes the S25 phosphorylation site and the N-terminal 20 Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA) was conducted 48 h amino acids are strongly predicted to be a disordered binding after iBAs were induced to differentiate. Purified DNA and P3000 reagent were region that also includes phosphorylation sites at S15 and S18. mixed with Opti-MEM. The mixture was then added into Opti-MEM containing Because the N-terminal 36 amino acids of Them1 are sufficient to Lipofectamine 3000. The three reagents were allowed to incubate at room tem- perature for 10 min and then added to the cells. After 7 h of transfection, the drive puncta formation in iBAs, our data strongly support that medium was replaced by maintenance media. After 48 h, the cells were assayed. this region is involved in puncta formation and should be further The same ratio of cells to DNA was used for each experiment but due to the very investigated. Alternatively, Them1 may function as a client. Cli- low transfection efficiency, it was not possible to assay Them1 concentration by ents often have molecular properties similar to scaffolds including immunoblotting. 20 For experiments using adenoviral vectors to express Them1, recombinant intrinsic disorder , also making this role for Them1 a distinct adenoviruses including Ad-Them1-EGFP, Ad-AAA-EGFP, Ad-DDD-EGFP, or possibility. Ad-EGFP were added to the medium for 24 h after day 1 of differentiation. Dissolution of biomolecular condensates can be used to change Adenoviral vectors were initially screened for MOI versus cell protein expression the available concentration of cellular proteins to affect reaction using our Them1 antibody11, mRNA expression using quantitative RT-PCR fi 19 (Supplementary Fig. 3b, c), and cell viability using the crystal violet assay26, which ef ciency , and is likely important for Them1 to reduce its local was nearly 100% with MOI’s ranging from 10 through 80 (not shown). Unless concentration at LD and mitochondria subsequently increasing otherwise specified, an MOI of 40 was used because Them1 expression was most fatty acid availability for β-oxidation in mitochondria and similar to that induced in BAT during cold exposure (Supplementary Fig. 3b). increasing thermogenesis. The phosphorylation state of proteins regulates their ability to assemble in biomolecular condensates, LC-MS/MS for detecting Them1 phosphorylation. Protein concentrations were with threonine phosphorylation resulting in condensation and estimated with a Bradford protein assay prior to acetone precipitation, done as described27. Pellets were resuspended in 8 M urea (Proteomics grade, PlusOne; GE serine phosphorylation reducing retention in a dose-dependent fi 16 Healthcare, Chicago IL), 5 mM TCEP (Bond-breaker; Thermo Fisher Scienti c, manner . Phosphorylation events were shown to regulate phase Waltham, MA), 50 mM triethyl ammonium bicarbonate (TEAB, Sigma-Aldrich) separation, aggregation, and the retention of mRNA’s in FUS and then alkylated with 100 mM iodoacetamide to a final concentration of 9 mM. proteins, which are neuronal inclusions of aggregated RNA- Samples were then diluted with 100 mM TEAB (Sigma-Aldrich) followed by binding protein fused in sarcoma23,24. It is thus phosphorylation incubation with trypsin (Sigma-Aldrich) at a ratio 1 : 40 trypsin:total protein overnight at 37 °C. Samples were acidified to a final concentration 0.1% TFA at S15 and S25, which are highly conserved residues in Them1 (Sigma-Aldrich) and then desalted using HLB solid phase extraction plates that reduces its retention in puncta as evidenced in inhibitor (Waters, Milford, MA). Eluted peptides were dried in a centrifugal vacuum con- studies and by genetic manipulation of Them 1. Norepinephrine- centrator and then stored at −20 °C. induced protein phosphorylation events during cold exposure in Phosphorylated peptides were enriched with Fe-NTA magnetic beads (Cube Biotech, Cambridge, MA) using a 96-well plate magnet28. Beads were washed and addition to a 3-fold increase in Them1 protein concentration over the phosphorylated peptides eluted with 50% acetonitrile and 1% ammonium 11 time , likely creates equilibrium between active Them1 in puncta hydroxide. Eluted peptides were then acidified with formic acid and dried. Before and inactive Them1 that is diffusely localized in the cytoplasm. LC-MS/MS analysis, samples were resuspended to yield a 0.3 μg/μL solution, Our data thus suggest that the delicate balance of Them1 protein assuming a 1 : 100 reduction in peptide amount. LC-MS/MS analysis was conducted with a 1200-bar nano HPLC (Thermo Easy concentration, its phosphorylation status, and environmental 1200 nLC; Thermo Fisher Scientific) using a 15 cm × 75 μm ID, 1.6 μm C18 column factors that promote Them1 phase transitions dictate the rate of (Ionoptickcs, Aurora series emitter; Fitzroy VIC, Australia). The temperature of the thermogenesis in BAT. column was maintained with an integrated column oven (PRSO-V1; Sonation lab

NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x solutions, Biberach, Germany). Enriched phosphorylated peptides were Treatment of iBAs to explore metabolic pathways. To explore the metabolic chromatographed using a mixed linear gradient of buffer A containing 0.1% formic pathway leading to reorganization of intracellular Them1, various treatments were acid in water and buffer B containing 0.1% formic acid in 80% acetonitrile. Flow applied for 4 h to cultured iBAs as follows: Norepinephrine (1 μM; Sigma Aldrich; rates were maintained over the course of the HPLC run and the gradient applied St. Louis, MO), a neurotransmitter, was used to mimic cold exposure. Forskolin (1 was (Time:%B): 0:3, 1:7, 81:30, 91:40, 93:80, 96:80, 97:3. The LC system was μM; Selleckchem, Houston, TX), a membrane permeable labdane diterpene pro- interfaced to an Eclipse Tribrid Orbitrap mass spectrometer (Thermo Fisher duced from the Coleus plant, was used to activate adenylyl cyclase. This treatment, Scientific). The positive spray voltage was set to 2–2.5 kV, ion transfer tube temp through the activation of cAMP, activates PKA. PMA (3 μM; LC Laboratories, 305 °C, RF funnel 30%. The FAIMS Pro system was set to survey three Woburn, MA) was used to activate PKC. NE, adenylyl cyclase, and PKC were compensation voltages of −40, −60, and − 80 V. The full profile MS spectra were selected for activation as they represent three key control points in the pathway collected at 120k resolution (200 m/z). The data-dependent acquisition used a allowing the characterization of all upstream regulators. Pathway inhibitors were monoisotopic peak determination filter set to peptides, intensity set to 5e3, charge added 1 h prior to activation and were continued throughout the 4 h activation states 2–6, dynamic exclusion of 60 s and a cycle time of 0.6 s for each FAIMS CV period as follows: PKI [14–22] myristoylated (0.5 μM; Invitrogen, Camarillo, CA), a used, total cycle time 1.8 s. The MS/MS fragmentation was collected with high synthetic peptide inhibitor of PKA, was added to inhibit PKA activation. Atglistatin energy collision induced dissociation with the following settings; quadrupole (40 μM; Selleckchem, Houston, TX) was used to selectively inhibit the processing of isolation with 1.6 Da window, normalized collision energy set to 30%, detector set triacylglycerol to fatty acids via the formation of diacylglycerol by ATGL. to ion trap with rapid scan rate and defined mass range of 200–1500, AGC 250% Ruboxistaurin, or LY333531, (2 μM; Selleckchem, Houston, TX), an isozyme- with a maximum injection time set to 35 ms data stored as centroid data. selective inhibitor of PKC that competitively and reversibly inhibits PKC-βI and A pooled sample of enriched peptides from all of the samples was crudely PKC-βII, was added to inhibit PKCβ activity. fractionated with centrifugal columns following manufacturer’s instructions (Pierce High pH reversed-phase peptide fractionation kit (Thermo Fisher Scientific). This fl generated 8 fractions, a ow through, and wash fraction all were analyzed using the Correlative light and electron microscopy of Them1 in iBAs. Differentiated iBAs same method as the samples. The results were used to generate an in-depth library transfected with a plasmid containing Them1 linked to APEX2 were fixed and the of phosphorylated peptides. APEX2 developed using an ImPACT peroxidase substrate kit (Vector Labs, Bur- Data were processed using Proteome Discoverer v2.4 (Thermo Fisher fi lingame, CA). Cells were then frozen using a Wohlwend Compact 02 High Pres- Scienti c). Data-dependent data were searched with Sequest module using the sure Freezer. Frozen cells underwent super-quick freeze substitution and following parameters; MS1 mass accuracy 10 ppm with MS/MS tolerance 0.6 Da, embedding using the protocol established by McDonald29. In 0.5-μm thick sec- searched against the Mus musculus protein database Uniprot taxonomy ID 10090 tions, positive cells stained brown by light microscopy. When positive cells were v2017-10-25, trypsin set as the digestion with maximum of three missed fi fi identi ed, four consecutive levels of serial ultra-thin sections were obtained with cleavages. Dynamic modi cation included oxidation of methionine, one thick section for light microscopy between levels. Light microscopy images, phosphorylation of Ser and Thr, N-terminal pyro-glutamate formation, N-terminal fi taken with an Axioimager (Zeiss) equipped with a color ccd camera were used to Met loss and Met loss with acetylation, static modi cation was set to follow puncta in the sections and to determine the position of positive cells in ultra- cabamidomethyl of Cys. Validation was done using the percolator processing node fi thin sections. Images were taken on a JEOL 1400 transmission electron microscope with FDR set to 0.01. Them1 (Q8VHQ9) phosphorylated peptides identi ed via (TEM) equipped with a Gatan (Pleasanton, CA) Orius SC1000 camera. Low Sequest were imported to Skyline for chromatographic extraction and integration. magnification TEM images were taken at an original magnification of ×4000–5000, The signal intensity of Them1 peptides was normalized to the phosphorylated intermediate magnification TEM images at ×25000, and high magnification TEM peptide RSS(phospho)QGVLHMPLYTSPIVK from isoform 2 of hormone- images at ×60,000. sensitive lipase (P54310-2). This peptide was chosen as a housekeeping phosphorylated peptide. The total amount of Them1 phosphorylation was normalized to this peptide. O2 consumption rates in iBAs. Oxygen consumption rates (OCR) were measured in iBAs using a Seahorse XFe96 (Agilent Technologies, Santa Clara, CA, USA). Fixation and Oil Red O staining for LD. Differentiated iBAs were washed with Briefly, iBAs were seeded at a density of 1000 cells/well in collagen-coated Seahorse phosphate buffered saline (PBS) and then fixed with 4% paraformaldehyde in PBS XF96 cell culture microplates 2 days before induction. On day 1 after induction, for 20 min at room temperature. Cells were washed twice before staining with 0.3% iBAs were incubated with Ad-Them1-EGFP (MOI 40), Ad-AAA-EGFP (MOI 50), Oil Red O solution in 60% isopropanol (Sigma Aldrich) at room temperature. The Ad-DDD-EGFP (MOI 60), or Ad-EGFP (MOI 40). In preliminary experiments cells were washed with PBS and then mounted with ProLong Diamond Antifade these MOIs led to similar Them1 and EGFP protein expression. One day after Mountant with DAPI (ThermoFisher) to identify nuclei. adenovirus infection, the medium was changed to the maintenance medium sup- μ plemented with 1 M NE. After 2 days, CO2 was withdrawn from the cultures for 1 h at 37 °C in Krebs-Henseleit buffer (pH 7.4) containing 2.5 mM glucose, 111 mM Immunostaining for LD, mitochondria, and plasmid-expressed Them1. Dif- fi NaCl, 4.7 mM KCl, 2 mM MgSO4-7H2O, 1.2 mM Na2HPO4, 5 mM HEPES, and ferentiated iBAs were xed, as above, and then stained for perilipin (anti-Plin1) to 0.5 mM carnitine (denoted as KHB). Basal OCR was measured in KHB for 18 min identify LD, Tomm20 (anti-Tomm20) to detect mitochondria, or anti-EGFP to μ fi fi and then 10 M NE was injected through the Seahorse injection ports to a nal visualize the plasmid-transfected Them1-EGFP. For staining Them1 in paraf n concentration of 1 μM in each well. NE responses were measured after injections as sections of BAT, tissues were treated with Image-iT™ FX Signal Enhancer (Thermo 11 fi previously described . Fisher Scienti c, Hampton, NH) after antigen retrieval with citrate buffer, pH 6.0. Relative transfection efficiency was assessed by counting the number of EGFP The sections were stained with anti-Them1 antibody, anti-Plin1, and Hoechst expressing cells relative to the number of total viable cells. Data were normalized to 33342 to identify nuclei. the number of viable cells, determined using the NucRed™ Live 647 ReadyProbes™ fi fi 11 Details of the antibodies: af nity puri ed rabbit anti-Them1 1:250 for assay (Thermo Fisher Scientific, Waltham, MA, USA) and the color read using a immunostaining or 1:1000 for immunoblots (see Supplementary Fig. 5); anti- Spectramax i3x (Molecular Devices, San Jose, CA, USA) visualized and calculated Tomm20, Santa Cruz Biotech (Dallas TX, USA), rabbit, sc-11415, 1:200; anti-Plin1, using the Spectramax i3x. Fitzgerald (Acton MA, USA), guinea pig, 20R-PP004,1:500; anti-EGFP, Novus Biologicals (Littleton CO, USA), goat, NB100-1678, 1:500; anti-actin, Abcam (Cambridge, MA, USA), mouse, Ab3280,1:1000. Secondary antibodies labeled with HRP, AlexaFluor 488, 647, or Cy3 were purchased from Jackson ImmunoResearch Fatty acid oxidation. Rates of oxidation of triglyceride-derived fatty acids were 30 (West Grove PA, USA). determined by pulse-chase methods as described by Cooper et al. . In brief, iBAs were cultured, differentiated, and transfected in 12-well plates as described above. On day 2 prior to pulse, the media was changed to pre-labeling medium including Confocal microscopy. The localization of fluorescence signal in cultured iBAs was DMEM, 10% FBS, 1 g/l glucose, and 10 mM HEPES. Cells were pulsed for 1.5 h evaluated using a Zeiss LSM880 confocal microscope system with or without with pre-labeling media containing 1 mM carnitine and 500 µM sodium oleate Airyscan. Images were acquired at high resolution as 2 µm z-stack slices through (Sigma-Aldrich) containing 2 µCi/ml [9,10-3H]oleate (Perkin Elmer, Waltham, the thickness of cells and assembled in 3-D using Volocity image processing MA) conjugated to BSA31. Cells were chased by washing with PBS containing 1% software. fatty acid-free BSA and then incubated for 1.5 h with pre-labeling media containing For confocal imaging of living cells, a time-lapse image was created to capture 1 mM carnitine, 500 µM oleate, and 5 µM NE. At the end of the chase period, cells the dynamics of protein expression and localization, signal transduction, and were washed with PBS containing 1% fatty acid-free BSA prior to analysis. Relative enzyme activity. Cells grown on coverslips were placed in the incubator chamber of transfection efficiency was determined by immunoblot analysis of Them1 a Zeiss LSM880 inverted confocal microscope system to visualize EGFP expression and EGFP. (linked to Them1). The interior of the chamber was kept at a constant 37 °C, and were extracted using the Folch method and then separated by thin-layer set to a 5% CO2 level and a 95% O2 level. An image was captured at T0, and chromatography using hexane:ethyl-ether:acetic acid (80:20:1, v/v/v) to determine immediately following the application of a treatment. The microscope then radiolabeled triglycerides31. The concentrations of total triglycerides were automatically tracked, focused, and photographed cells every 30 min for 4–5h. determined enzymatically (Wako Diagnostics, Mountain View, CA)32. Rates of These images are also taken in the high resolution z-stack format and analyzed fatty acid oxidation were determined from utilization of triglycerides during the using Volocity software. chase period and expressed relative to Ad-EGFP.

12 NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x ARTICLE

Activation of BAT in mice. Male C57BL/6J mice (7 w) were purchased from the References Jackson Laboratory (Bar Harbor, ME) and housed in a specific pathogen-free 1. Benador, I. Y. et al. Mitochondria bound to lipid droplets have unique AAALAC International Facility with a standard 12 h alternate light/dark cycle at an bioenergetics, composition, and dynamics that support lipid droplet ambient temperature of 22 ± 2 °C and 30–70% humidity, at Weill Cornell Medical expansion. Cell Metab. 27, 869–885 (2018). College. Animal use and euthanasia protocols were approved by the Institutional 2. Rambold, A. S., Cohen, S. & Lippincott-Schwartz, J. Fatty acid trafficking in Animal Care and Use Committee at Weill Cornell Medical College. starved cells: regulation by lipid droplet lipolysis, autophagy, and For the experiment, mice were moved to Promethion metabolic cages (Sable mitochondrial fusion dynamics. Dev. Cell 32, 678–692 (2015). Systems International, Las Vegas, NV) with one mouse/cage. Mice were housed at 3. Cannon, B. & Nedergaard, J. Brown adipose tissue: function and physiological fi 22 °C for the rst 24 h, but were changed to 30 °C overnight prior to experiments. significance. Physiol. Rev. 84, 277–359 (2004). Room temperature, or 22 °C, is below the thermoneutral zone for mice and the 4. Deiuliis, J. A. et al. β -adrenergic signaling acutely down regulates adipose 11 3 mild cold exposure stimulates Them1 expression in BAT . Experimental mice triglyceride lipase in brown adipocytes. Lipids 45, 479–489 (2010). were injected subcutaneously with one dose of CL316,243 (Sigma-Aldrich, 1 mg/kg 5. Brasaemle, D. L. Thematic review series: adipocyte biology. The perilipin β body weight) in saline, which is a 3 adrenergic receptor agonist. Control mice family of structural lipid droplet proteins: stabilization of lipid droplets and were injected with saline alone. Subscapular BAT was excised from mice after – fi control of lipolysis. J. Lipid Res. 48, 2547 2559 (2007). euthanasia at T0, 1, 2, or 4 h after injection. Tissues were xed overnight in 10% 6. Ellis, J. M. et al. Adipose acyl-CoA synthetase-1 directs fatty acids toward beta- neutral buffered formalin and then processed. Paraffin sections were used for H&E oxidation and is required for cold thermogenesis. Cell Metab. 12,53–64 stained samples and for immunocytochemistry as described above. (2010). 7. Grevengoed, T. J., Klett, E. L. & Coleman, R. A. Acyl-CoA metabolism and Image processing and quantification of fluorescence signal. For experiments partitioning. Annu. Rev. Nutr. 34,1–30 (2014). aimed to quantify changes in intercellular Them1 localization after stimulation, 8. Tillander, V., Alexson, S. E. H. & Cohen, D. E. Deactivating fatty acids: acyl- image segmentation was used to calculate the volume of Them1 by thresholding CoA thioesterase-mediated control of . Trends Endocrinol. the green (EGFP) pixels of specified intensity that represented Them1-EGFP minus Metab. 28, 473–484 (2017). background intensity using either Volocity software (Quorum Technologies, 9. Adams, S. H. et al. BFIT, a unique acyl-CoA thioesterase induced in Ontario, Canada) or Imaris software (Oxford Instruments, Concord, MA, USA). thermogenic brown adipose tissue: cloning, organization of the human gene The strategy used for both iBAs and for tissue Them1 expression was that puncta and assessment of a potential link to obesity. Biochem. J. 360, 135–142 (2001). represent concentrated Them1, which occupies a small intracellular volume 10. Zhang, Y. et al. Targeted deletion of thioesterase superfamily member 1 whereas diffuse Them1 is distributed evenly throughout the entire cell volume promotes energy expenditure and protects against obesity and insulin resulting in a large intracellular volume (Fig. 1k, schematic diagram). Them1 resistance. Proc. Natl Acad. Sci. USA 109, 5417–5422 (2012). fluorescence signal, which represented pixel intensity in the EGFP (green) channel, – 11. Okada, K. et al. Thioesterase superfamily member 1 suppresses cold was measured by using a normalized histogram at each 12-bit (0 4096) intensity thermogenesis by limiting the oxidation of lipid droplet-derived fatty acids in value. For iBAs, background signal was determined from cells that were not brown adipose tissue. Mol. Metab. 5, 340–351 (2016). transfected and the fluorescence threshold set to exclude this low-intensity back- 12. Huttlin, E. L. et al. A tissue-specific atlas of mouse protein phosphorylation ground signal. For paraffin sections of brown adipose tissue, the background and expression. Cell 143, 1174–1189 (2010). fluorescence intensity was determined using Them1-deficient mice and green pixels 13. Han, S. & Cohen, D. E. Functional characterization of thioesterase superfamily in this range were excluded from the experimental data. This strategy did not member 1/acyl-CoA thioesterase 11: implications for metabolic regulation. J. stratify the signal into intense (as found in puncta) or weak (as found in the diffuse – state), but instead identified the volume of total intracellular Them1 fluorescence Lipid Res. 53, 2620 2631 (2012). irrespective of intensity. 14. Eichmann, T. O. et al. Studies on the substrate and stereo/regioselectivity of adipose triglyceride lipase, hormone-sensitive lipase, and diacylglycerol-O- acyltransferases. J. Biol. Chem. 287, 41446–41457 (2012). Bioinformatics. The amino acid sequence of Them1 from Mus musculus was 15. Friend, D. S. & Brassil, G. E. Osmium staining of endoplasmic reticulum and 33 acquired from UniProt and analyzed with multiple servers. The IUPRED2A mitochondria in the rat adrenal cortex. J. Cell Biol. 46, 252–266 (1970). 34–37 server was used to predict intrinsic disorder and ANCHOR binding regions . 16. Li, P. et al. Phase transitions in the assembly of multivalent signaling proteins. The Prion-like Amino Acid Composition (PLAAC) server was used to predict – 38 Nature 483, 336 340 (2012). prion-like regions through the prion-like domains (PLD) and FOLD curves . The 17. Pak, C. W. et al. Sequence determinants of intracellular phase separation by propensity of Them1 to phase separate through long-range planar pi-pi contacts complex coacervation of a disordered protein. Mol. Cell 63,72–85 (2016). was predicted using a server generated by the laboratory of Professor Julie Forman- 18. Bansode, R. R., Huang, W., Roy, S. K., Mehta, M. & Mehta, K. D. Protein Kay (PScore)39. The Motif Scan server (https://myhits.isb-sib.ch/cgi-bin/ kinase C deficiency increases fatty acid oxidation and reduces fat storage. J. motif_scan) was utilized to examine the first 20 amino acids for known motifs40. Biol. Chem. 283, 231–236 (2008). 19. Alberti, S., Gladfelter, A. & Mittag, T. Considerations and challenges in Statistics and reproducibility. Data were analyzed using SigmaPlot software with studying liquid-liquid phase separation and biomolecular condensates. Cell a P value < 0.05 considered significant. Prior to analysis, data was subjected to an 176, 419–434 (2019). outlier test (Grubbs test; GraphPad on-line calculator https://www.graphpad.com/ 20. van Leeuwen, W. & Rabouille, C. Cellular stress leads to the formation of quickcalcs/grubbs1/) and individual data points were excluded that were extreme membraneless stress assemblies in eukaryotic cells. Traffic 20, 623–638 (2019). outliers, alpha value of 0.01 or less. For comparison of data from many groups, 21. Martin, E. W. & Mittag, T. Relationship of sequence and phase separation in fi one-way analysis of variance was used to determine statistical signi cance. If protein low-complexity regions. Biochemistry 57, 2478–2487 (2018). variances were not normal, non-parametric ANOVA was performed using ranks. 22. Letunic, I. & Bork, P. 20 years of the SMART protein domain annotation For experiments with multiple times and treatments, 2-way analysis of variance resource. Nucleic Acids Res. 46, D493–D496 (2018). fi was used to determine statistical signi cance. Post-hoc analyses were done to 23. Han, T. W. et al. Cell-free formation of RNA granules: bound RNAs identify determine differences between groups. features and components of cellular assemblies. Cell 149, 768–779 (2012). 24. Monahan, Z. et al. Phosphorylation of the FUS low-complexity domain Reporting summary. Further information on research design is available in the Nature disrupts phase separation, aggregation, and toxicity. EMBO J. 36, 2951–2962 Research Reporting Summary linked to this article. (2017). 25. Uldry, M. et al. Complementary action of the PGC-1 coactivators in Data availability mitochondrial biogenesis and brown fat differentiation. Cell Metab. 3, 333–341 (2006). The authors declare that the main data supporting the findings of this study are available fi 26. Seo, J. H., Fox, J. G., Peek, R. M. & Hagen, S. J. N-methyl D-aspartate channels within the article and its Supplementary Information les. Adenoviral vectors or plasmid link ammonia and epithelial cell death mechanisms in Helicobacter pylori constructs are available upon reasonable request to the corresponding authors. The raw – fi infection. Gastroenterology 141, 2064 2075 (2011). data les for LSMS were deposited at the Center for Computational Mass Spectrometry, 27. Nickerson, J. L. & Doucette, A. A. Rapid and quantitative protein precipitation Center for Computer Science and Engineering, University of California, San Diego. The for proteome analysis by mass spectrometry. J. Proteome Res. 19, 2035–2042 dataset files can be accessed under the dataset ID MSV00008675 [https://massive.ucsd. = (2020). edu/ProteoSAFe/dataset.jsp?task cc447e32ff414e9f8002deaa8c5566eb]. Source data are 28. Searle, B. C. et al. Chromatogram libraries improve peptide detection and provided with this paper. quantification by data independent acquisition mass spectrometry. Nat. Commun. 9, 5128 (2018). Received: 12 March 2020; Accepted: 29 April 2021; 29. McDonald, K. L. Rapid embedding methods into epoxy and LR White resins for morphological and immunological analysis of cryofixed biological specimens. Microsc. Microanal. 20, 152–163 (2014).

NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications 13 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23595-x

30. Cooper, D. E., Grevengoed, T. J., Klett, E. L. & Coleman, R. A. -3- Author contributions phosphate acyltransferase isoform-4 (GPAT4) limits oxidation of exogenous Y.L. did the cell culture experiments and all confocal microscopy experiments; N.I. did fatty acids in brown adipocytes. J. Biol. Chem. 290, 15112–15120 (2015). OCR studies without PMA; H.T.N. and T.I.K. performed the animal experiments, H.T.N. 31. Alves-Bezerra, M. et al. Thioesterase superfamily member 2 promotes hepatic also conducted OCR studies with PMA and lipid oxidation experiments; B.R.R. and VLDL secretion by channeling fatty acids into triglyceride . A.M.R. conducted the LC-MS/MS studies; M.B. assisted with experiments; L.H.A. did the Hepatology 70, 496–510 (2019). immunostaining; M.T. and E.O. did the Them1 bioinformatic analysis; D.E.C. co- 32. Nicholls, H. T., Hornick, J. L. & Cohen, D. E. Phosphatidylcholine transfer mentored Y.L., assisted with experimental design, and edited the paper. S.J.H. did CLEM, protein/StarD2 promotes microvesicular steatosis and liver injury in murine co-mentored Y.L., assisted with experimental design, and wrote the paper. All authors experimental steatohepatitis. Am. J. Physiol. Gastrointest. Liver Physiol. 313, reviewed and edited the final manuscript. G50–G61 (2017). 33. UniProt Consortium UniProt: a worldwide hub of protein knowledge. Nucleic Competing interests Acids Res. 47, D506–D515 (2019). 34. Meszaros, B., Erdos, G. & Dosztanyi, Z. IUPred2A: context-dependent The authors declare no competing interests. prediction of protein disorder as a function of redox state and protein binding. Nucleic Acids Res. 46, W329–W337 (2018). Additional information 35. Dosztanyi, Z. Prediction of protein disorder based on IUPred. Protein Sci. 27, Supplementary information The online version contains supplementary material 331–340 (2018). available at https://doi.org/10.1038/s41467-021-23595-x. 36. Dosztanyi, Z., Csizmok, V., Tompa, P. & Simon, I. The pairwise energy content estimated from amino acid composition discriminates between folded Correspondence and requests for materials should be addressed to D.E.C. or S.J.H. and intrinsically unstructured proteins. J. Mol. Biol. 347, 827–839 (2005). 37. Meszaros, B., Simon, I. & Dosztanyi, Z. Prediction of protein binding regions Peer review information Nature Communications thanks the anonymous reviewers for in disordered proteins. PLoS Comput. Biol. 5, e1000376 (2009). their contributions to the peer review of this work. Peer review reports are available. 38. Lancaster, A. K., Nutter-Upham, A., Lindquist, S. & King, O. D. PLAAC: a web and command-line application to identify proteins with prion-like amino Reprints and permission information is available at http://www.nature.com/reprints acid composition. Bioinformatics 30, 2501–2502 (2014). 39. Vernon, R. M. et al. Pi-Pi contacts are an overlooked protein feature relevant Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in to phase separation. Elife. 7, e31486 (2018). published maps and institutional affiliations. 40. Sigrist, C. J. et al. PROSITE, a protein domain database for functional characterization and annotation. Nucleic Acids Res. 38, D161–D166 (2010). Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, Acknowledgements adaptation, distribution and reproduction in any medium or format, as long as you give The authors thank Dr. Bruce Spiegelman, Dana-Farber Cancer Institute, Harvard appropriate credit to the original author(s) and the source, provide a link to the Creative Medical School for providing the iBAs, Mr. Kyle Smith for his technical help with Commons license, and indicate if changes were made. The images or other third party electron microscopy, and Mr. Aniket Gad for expert advice on image processing. This material in this article are included in the article’s Creative Commons license, unless work was supported by the National Institutes of Health (RO1 DK 103046 to D.E.C., indicated otherwise in a credit line to the material. If material is not included in the S.J.H., and E.O.; R01 DK048873 to D.E.C.), the Harvard Digestive Diseases Center (P30 article’s Creative Commons license and your intended use is not permitted by statutory DK034854 to S.J.H.), and the National Institutes of Health shared-instrumentation grant regulation or exceeds the permitted use, you will need to obtain permission directly from program for the High Pressure Freezer (S10 OD019988-01 to S.J.H.). H.T.N is the the copyright holder. To view a copy of this license, visit http://creativecommons.org/ recipient of a Pinnacle Research Award from the AASLD Foundation and T.I.K. is the licenses/by/4.0/. recipient of a Weill Cornell Department of Medicine Pre-Career Award and acknowl- edges support from NIH T32DK007533. N.I. is the recipient of an America Heart Association Postdoctoral Fellowship and S.G. was supported by a Research Science © The Author(s) 2021 Institute/Center for Excellence in Education Summer Research Fellowship.

14 NATURE COMMUNICATIONS | (2021) 12:3493 | https://doi.org/10.1038/s41467-021-23595-x | www.nature.com/naturecommunications